Answer customers from your own documents, give your developers a bot that can safely reach real systems, and hand any conversation to a human without losing the thread.
Bring your own AI model — OpenAI, Anthropic, Gemini, Deepseek, or a local one.
Upload PDFs, Word files, Markdown or a web page and the bot answers from them. Each bot has its own isolated knowledge base, so one bot can never quote another’s documents. When it does not know, it says so instead of inventing an answer.
A customer can ask for a person, or the bot can escalate itself when its confidence drops. The conversation appears in a live inbox, the bot pauses, and your agent replies through the same Telegram chat. Close it and the bot picks up again.
Read logs over SSH, query a database, check a repository, run a registered deploy script. Every write, delete or deploy stops and shows you the exact command for approval first, and anything aimed at production always asks.
A deterministic rule layer sits underneath the model’s own judgement and force-flags destructive commands. Tools are granted per bot in the execution layer, so a bot that was never given a tool cannot use it however the conversation goes. Every tool run is written to an append-only audit log.
The same bot answers in a chat widget you drop onto your site with one script tag. Origins are allow-listed per widget, so nobody else can embed your bot and spend your quota.
Messages, active users, tokens and estimated cost per bot per day. Set a daily ceiling and the platform pauses itself rather than surprising you. Per-user rate limits and a one-click kill switch come as standard.
Pick a plan and your workspace is ready in a minute. No card details are handled by us — payment goes through Stripe.
See pricing